Previously, most peripheral blood-based DNA methylation studies used genomic DNA isolated from whole blood or peripheral blood mononuclear cells (PBMC) to investigate the methylation signatures of COVID-19 and their relationship with clinical severities and outcomes [18,19,20,21,22], while the significance of cfDNA methylation in COVID-19 progression has not been fully understand [11, 12, 23]. This study comprehensively described the distinct cfDNA methylation features related to COVID-19, and evaluated cell death events related to lung and immune cells in COVID-19 through genome-wide DNA methylation sequencing of COVID-19 patients and healthy controls. The elevated cfDNA derived from for alveolar epithelial cells was correlated with clinical severities and outcomes in patients with COVID-19.

Our study observed that the global cfDNA methylation levels decreased with the increasing of disease severity, and the methylation pattern of severe patients obviously differed from non-hospitalized and healthy cases. These findings suggested that patients with severe disease had more abnormalities of methylation and more uncontrolled gene expression. DMR analyses among three groups showed that there were more hypo-methylated DMRs in the two COVID-19 groups, and there were significantly higher number of hypo-methylated DMRs than hyper-methylated DMRs in the promoter region for the comparison of severe and non-hospitalized cases, indicating that more genes in the case group were hypo-methylated, leading to abnormal activation of gene expression. These findings were consistent with the previous work conducted by Balnis et al., revealing higher numbers of hypo-methylated DMRs than hyper-methylated DMRs by comparing the cfDNA methylation profiles of 15 patients hospitalized with COVID-19 and 15 healthy volunteers [23].

In addition, we found that severe COVID-19 had distinct cfDNA methylation profiles compared with non-hospitalized cohort with the identification of 11,156 DMRs and enrichment of 286 pathways. Most of the top ten pathways were involved in immune response. It is reasonable in terms of the defense between immune system and COVID-19. Patients with severe COVID-19 experience the coexistence of immunosuppression and hyperinflammation states, characterized by the decrease of lymphocytes and elevation of inflammatory cytokines, respectively [24]. Compared with non-severe cases, severe cases tend to have lower lymphocytes and higher leukocytes, infection-related biomarkers and inflammatory cytokines [4]. The DMRs between severe and non-hospitalized cases were enriched in neutrophil activation and degranulation, T cell activation and leukocytes cell–cell adhesion, inflammatory response regulation and myeloid cell differentiation pathways, which was consistent with the difference of immune response between severe and non-severe patients [4, 24, 25]. Multiple studies confirmed the connection between GTPase signaling regulation and acute respiratory distress syndrome (ARDS) and subsequent pulmonary fibrosis [26], which was consistent with our GO pathway enrichment findings.

The tissue fraction for immune cells obtained by cfDNA methylation sequencing reflects cell death events of immune cells in vivo rather than the number of immune cells in blood [27]. The death of immune cells may result from the drastic immune response that they participate in, or lymphopenia caused by virus infection. In the early stage of COVID-19 infection (usually within a few hours as other virus’ infection), tissue-resident innate immune cells, particularly neutrophils and monocytes in the nasopharyngeal mucosa, are stimulated by chemokines. They initially recognize the viral infection and recruit more innate immune cells to eliminate the virus. Adaptive immunity, including B cells and T cells, starts participating in the antiviral process a few days after infection [28]. Most patients (including asymptomatic and mild COVID-19 patients) can clear the virus, and the counts of lymphocytes as well as immune response will gradually return to normal levels. While severe cases experience a hyperinflammation phase, accompanied by persistently lower levels of lymphocytes called lymphopenia [24]. In our study, plasma samples were collected 7 days post-infection in the non-hospitalized patients. Adaptive immunity is likely playing a dominant role at this time point. Thus, it was reasonable that active adaptive immune response may result in more cell death of B cells and thus elevated cfDNA methylation markers of B cells. In contrast, severe COVID-19 patients who had blood collected immediately after ICU admission were undergoing a phase of hyperinflammation and lymphopenia. Therefore, the cfDNA methylation markers of B cells, T cells, and granulocytes were all elevated due to increased cell death of these cells. In addition, previous studies observed the accumulation of more natural killer cells in mild COVID-19 rather than severe COVID-19, suggesting the potential role of natural killer cells to prevent over-inflammation and tissue injury [29,30,31,32,33], which may explain the reduced cfDNA from natural killer cells in severe COVID-19 compared with non-hospitalized patients. Plasma cfDNA derived from immune cells provides a novel biomarker to for monitoring immune responses in patients with COVID-19.

A panel of lung-specific methylation markers, targeting alveolar and bronchial epithelial cells of lung was applied to access lung-derived cfDNA in plasma samples from healthy individuals, patients with lung cancer, and patients with chronic obstructive pulmonary disease. The study revealed that normal lung cell turnover likely releases cfDNA into the air spaces, rather than to the bloodstream [34]. Lung-derived cfDNA is observed in the plasma when there is a pathological disruption of lung tissue architecture, as seen in lung cancer and to a lesser extent in other lung diseases [34]. Significant lung injury was observed in COVID-19 patients in our study, indicating that lung damage in COVID-19 is sufficient to reverse tissue topology and release cfDNA to blood rather than to the air spaces.

We also found that as the disease worsened, the lung epithelial cells were more severely damaged, especially for the alveolar epithelial cells. Tissue fraction for alveolar epithelial cells had the best performance to distinguish severe cases from non-hospitalized cases. These findings were in accordance with the respiratory epithelial cell responses to SARS-CoV-2. In conducting airways, ciliated cells and secretory cells are the main cells infected and their response to interferon and cytokine is moderate after SARS-CoV-2 infection. The alveolar epithelial cell is the main target cell type in the gas exchange portion of the lung. Alveolar cell death and marked innate immune response during infection likely impede epithelial repair mechanisms and contribute to alveolar damage and resultant acute respiratory distress syndrome [35].

Two previous studies [11, 12], respectively, conducted whole genome bisulfite sequencing of plasma cfDNA from COVID-19 patients. Based on the publicly available reference methylation atlas of human tissues, they deconvoluted the tissue origins using CelFiE algorithm and a non-negative least-squares method (not specified), respectively. Then, they multiplied the relative estimated proportions of cfDNA for lung by the total concentration of cfDNA in plasma to ascertain the absolute cfDNA concentration of lung. In Andargie et al.’s study [11], lung cfDNA level was correlated with COVID-19 disease severity and outcome, showing an AUC of 0.938 and 0.847, respectively. It is noteworthy that samples from Cheng et.al.'s research were sequenced to a minimum depth of 0.7 × human genome coverage. We speculated that this low sequencing depth might lead to inaccurate identification of low-abundance signals in cfDNA, and thus the AUC of 0.56 for lung cfDNA in predicting WHO ordinal scores [12]. The average sequencing depth in our study was 63x, and our study utilized cell type-specific methylation markers from Netanel et.al.'s research to pinpoint lung damage to alveolar epithelial cells [2], which could effectively differentiate COVID-19 cases with varying severities, lung injury levels, SOFA scores, and in-hospital deaths, achieving AUCs of 0.958, 0.941, 0.919, and 0.955, respectively. Therefore, our biomarker offers a more accurate insight into tissue injury and outperforms previous studies in its performance.

Several limitations needed to be disclosed in the study. Firstly, this study included a limited number of samples, especially for patients with severe COVID-19. The findings of this study need to be further validated in a large cohort study. In addition, all patients that transferred to our department already progressed to severe disease, and samples from moderate COVID-19 patients were not included in this study. Furthermore, the blood for severe and non-hospitalized patients was collected 17 days after symptom presentation and 7 days after diagnosis, respectively. It is more interesting to further explore the specific cfDNA methylation profile for COVID-19 patients at the initial stage of infection, and the performance of cfDNA from lung alveolar epithelial cell to predict COVID-19 severity, prognosis and death. Moreover, few patients in our study developed multi-organ dysfunction, which makes it impossible to investigate the relationship between cfDNA methylation signatures and injuries derived from other tissue or cell types. Plasma cfDNA methylation signatures may have potential applications in other settings, warranting future investigation.